Negative Control Outcome Research Articles

The Association Between Cancer and Spousal Rate of Memory Decline: A Negative Control Study to Evaluate (Unmeasured) Social Confounding of the Cancer-memory Relationship.

Cancer diagnoses are associated with better long-term memory in older adults, possibly reflecting a range of social confounders that increase cancer risk but improve memory. We used spouse's memory as a negative control outcome to evaluate this possible confounding, since spouses share social characteristics and environments, and individuals' cancers are unlikely to cause better memory among their spouses. We estimated the association of an individual's incident cancer diagnosis (exposure) with their own (primary outcome) and their spouse's (negative control outcome) memory decline in 3601 couples from 1998 to 2014 in the Health and Retirement Study, using linear mixed-effects models. Incident cancer predicted better long-term memory for the diagnosed individual. We observed no association between an individual's cancer diagnosis and rate of spousal memory decline. This negative control study suggests that the inverse association between incident cancer and rate of memory decline is unlikely to be attributable to social/behavioral factors shared between spouses.

Association between methylphenidate treatment and risk of seizure: a population-based, self-controlled case-series study

Individuals with attention-deficit hyperactivity disorder (ADHD) are at increased risk of seizures. Stimulant medications such as methylphenidate are the most commonly prescribed treatment for ADHD, but the association between their therapeutic use and the risk of seizures is unclear. We aimed to investigate the association between methylphenidate treatment and the risk of seizure. For this population-based observational study, we used the electronic medical record database of the Hong Kong Clinical Data Analysis And Reporting System to identify individuals aged 6-25 years who received at least one methylphenidate prescription during the study period. Individuals with records of seizure or epilepsy before the study period were excluded. Individuals treated with methylphenidate who had seizures during the study period were included in the subsequent analyses, and a self-controlled case-series design was used to control for time-invariant individual characteristics. We did additional analyses using skin infection as a negative control outcome. We compared relative incidence of seizure during periods when individuals were exposed to methylphenidate with that during non-exposed periods. Of 29 604 individuals prescribed methylphenidate between Jan 1, 2001, and Dec 31, 2017, 269 (199 males and 70 females) had incident seizures. The mean age at baseline was 6·66 years (SD 2·01) and the median age at the incident seizure was 9·69 years (IQR 7·62-12·99). The overall incidence of seizure during methylphenidate treatment was 4·4 per 10 000 patient-years. We detected an increased risk of seizure during the first 30 days of methylphenidate treatment compared with that during non-exposed periods, with an incidence rate ratio of 4·01 (95% CI 2·09-7·68). No increase in risk was identified during the following 31-180 days of treatment (1·13, 0·56-2·25) or during subsequent treatment (1·38, 0·92-2·07). We did not identify an increased risk in any risk window for the negative control outcome analysis. No individuals died because of a seizure during the study period. The incidence of seizures was higher in the period immediately after the start of methylphenidate treatment than in the non-exposed period. No increased risk was observed during continuation of methylphenidate treatment. The association between methylphenidate treatment and seizures immediately after initiation of medication can be seen as a potential safety signal. Monitoring of neurological outcomes in individuals with ADHD is recommended when they first start methylphenidate treatment. Hong Kong Research Grants Council.

Factors associated with adjuvant endocrine therapy adherence in non-metastatic breast cancer.

533 Background: Over 50% of breast cancer patients prescribed adjuvant endocrine therapy (ET) are nonadherent during the recommended 5-year course of therapy. We investigated the role of cancer medication delivery on adherence, including mail order pharmacy use, number of pharmacies and copays. Methods: We conducted a retrospective cohort study of 15,769 commercially insured breast cancer patients age 18-64 with newly diagnosed breast cancer in 2007-2015 that initiated ET. Incident breast cancer was identified by a validated algorithm which identifies mainly non-metastatic breast cancer. We examined the association between 12-month AET adherence (proportion of days covered by fills ≥0.80) and mail order pharmacy use, number of pharmacies, and AET copays. We used Poisson regression to estimate nonadherence risk ratios and adjusted for demographics (age, income, race, urbanicity), comorbidities, total medications, primary cancer treatments (surgery, radiation, chemo, and ET initiated), and generic AI availability. Sensitivity analyses were conducted using alternate specifications for independent variables. To test whether any observed differences were due to self-selection, we also conducted a negative control analysis. Results: Most patients were white (74.4%) and age 55-64 (43.3%). Only 16% of patients used a mail order pharmacy for ET fills, most patients only used one pharmacy (58.8%) and 25.2% had a co-pay of $20 or more. In the primary analysis, mail order patients were more likely to be adherent to their ET (aRR 1.21; 95% CI 1.18-1.24), patients using one pharmacy were more likely to be adherent (1 vs 3+: aRR 1.09; 95% CI 1.06-1.13), and patients with lower copays were more likely to be adherent (quartile 1 vs 4: aRR 1.04; 95% CI 1.01-1.08). Results were consistent across sensitivity analyses, and there was no association between mail order and copays and the negative control outcome of any pneumonia diagnoses. Conclusions: Medication delivery factors are associated with adherence to breast cancer AET. Future work should investigate whether interventions to streamline medication delivery could improve adherence for this population.

Prior event rate ratio adjustment produced estimates consistent with randomized trial: a diabetes case study

ObjectivesElectronic health records (EHR) provide a valuable resource for assessing drug side-effects, but treatments are not randomly allocated in routine care creating the potential for bias. We conduct a case study using the Prior Event Rate Ratio (PERR) Pairwise method to reduce unmeasured confounding bias in side-effect estimates for two second-line therapies for type 2 diabetes, thiazolidinediones, and sulfonylureas. Study Design and SettingsPrimary care data were extracted from the Clinical Practice Research Datalink (n = 41,871). We utilized outcomes from the period when patients took first-line metformin to adjust for unmeasured confounding. Estimates for known side-effects and a negative control outcome were compared with the A Diabetes Outcome Progression Trial (ADOPT) trial (n = 2,545). ResultsWhen on metformin, patients later prescribed thiazolidinediones had greater risks of edema, HR 95% CI 1.38 (1.13, 1.68) and gastrointestinal side-effects (GI) 1.47 (1.28, 1.68), suggesting the presence of unmeasured confounding. Conventional Cox regression overestimated the risk of edema on thiazolidinediones and identified a false association with GI. The PERR Pairwise estimates were consistent with ADOPT: 1.43 (1.10, 1.83) vs. 1.39 (1.04, 1.86), respectively, for edema, and 0.91 (0.79, 1.05) vs. 0.94 (0.80, 1.10) for GI. ConclusionThe PERR Pairwise approach offers potential for enhancing postmarketing surveillance of side-effects from EHRs but requires careful consideration of assumptions.

Effect of Initiating Cardiac Rehabilitation After Myocardial Infarction on Subsequent Hospitalization in Older Adults.

Outpatient cardiac rehabilitation (CR) participation after myocardial infarction (MI) reduces all-cause mortality; however, less is known about effects of CR on post-MI hospitalization. The study objective was to investigate effects of CR on hospitalization following acute MI among older adults. Medicare beneficiaries aged 65 to 88 yr hospitalized in 2008 with acute MI, who survived at least 60 d post-discharge, had a revascularization procedure during index hospitalization, and did not have an MI in previous year were eligible for this study. CR initiation was assessed in the 60 d post-discharge. Competing risk survival analysis was used to estimate the proportion of discharged beneficiaries hospitalized between the end of 60-d exposure window and December 31, 2009, treating death as a competing event. The mean ± SD age of 32 851 Medicare beneficiaries meeting study criteria was 75 ± 6.0 yr, approximately half were male (52%), and the majority were white (88%). In this study, 21% of beneficiaries initiated CR within the exposure window. At 1 yr post-discharge, CR initiators had a lower risk of recurrent MI (4.2% [95% CI, 3.5-5.1]), cardiovascular (15.7% [95% CI, 14.3-17.2]), and all-cause (30.4% [95% CI, 28.8-32.1]) hospitalization than noninitiators (5.2% [95% CI, 5.0-5.5]; 18.0% [95% CI, 17.6-18.4]; and 33.2% [95% CI, 32.5-33.8], respectively). There was no difference in fracture risk (negative control outcome). This study provides evidence that CR can reduce the 1-yr risk of cardiovascular and all-cause hospital admissions in Medicare aged MI survivors.

Effect of maternal smoking during pregnancy on childhood type 1 diabetes: a whole-of-population study.

Evidence of an association between maternal smoking during pregnancy (prenatal smoking) and childhood type 1 diabetes is mixed. Previous studies have been small and potentially biased due to unmeasured confounding. The objectives of this study were to estimate the association between prenatal smoking and childhood type 1 diabetes, assess residual confounding with a negative control design and an E-value analysis, and summarise published effect estimates from a meta-analysis. This whole-of-population study (births from 1999 to 2013, participants aged ≤15years) used de-identified linked administrative data from the South Australian Early Childhood Data Project. Type 1 diabetes was diagnosed in 557 children (ICD, tenth edition, Australian Modification [ICD-10-AM] codes: E10, E101-E109) during hospitalisation (2001-2014). Families not given financial assistance for school fees was a negative control outcome. Adjusted Cox proportional HRs were calculated. Analyses were conducted on complete-case (n = 264,542, type 1 diabetes = 442) and imputed (n = 286,058, type 1 diabetes = 557) data. A random-effects meta-analysis was used to summarise the effects of prenatal smoking on type 1 diabetes. Compared with non-smokers, children exposed to maternal smoking only in the first or second half of pregnancy had a 6% higher type 1 diabetes incidence (adjusted HR 1.06 [95% CI 0.73, 1.55]). Type 1 diabetes incidence was 24% lower (adjusted HR 0.76 [95% CI 0.58, 0.99]) among children exposed to consistent prenatal smoking, and 16% lower for exposure to any maternal smoking in pregnancy (adjusted HR 0.84 [95% CI 0.67, 1.08]), compared with the unexposed group. Meta-analytic estimates showed 28-29% lower risk of type 1 diabetes among children exposed to prenatal smoking compared with those not exposed. The negative control outcome analysis indicated residual confounding in the prenatal smoking and type 1 diabetes association. E-value analysis indicated that unmeasured confounding associated with prenatal smoking and childhood type 1 diabetes, with a HR of 1.67, could negate the observed effect. Our best estimate from the study is that maternal smoking in pregnancy was associated with 16% lower childhood type 1 diabetes incidence, and some of this effect was due to residual confounding.

Multiply robust causal inference with double-negative control adjustment for categorical unmeasured confounding.

Unmeasured confounding is a threat to causal inference in observational studies. In recent years, the use of negative controls to mitigate unmeasured confounding has gained increasing recognition and popularity. Negative controls have a long-standing tradition in laboratory sciences and epidemiology to rule out non-causal explanations, although they have been used primarily for bias detection. Recently, Miao and colleagues have described sufficient conditions under which a pair of negative control exposure and outcome variables can be used to identify non-parametrically the average treatment effect (ATE) from observational data subject to uncontrolled confounding. We establish non-parametric identification of the ATE under weaker conditions in the case of categorical unmeasured confounding and negative control variables. We also provide a general semiparametric framework for obtaining inferences about the ATE while leveraging information about a possibly large number of measured covariates. In particular, we derive the semiparametric efficiency bound in the non-parametric model, and we propose multiply robust and locally efficient estimators when non-parametric estimation may not be feasible. We assess the finite sample performance of our methods in extensive simulation studies. Finally, we illustrate our methods with an application to the post-licensure surveillance of vaccine safety among children.

Association between beta-blocker use and mortality/morbidity in older patients with heart failure with reduced ejection fraction. A propensity score-matched analysis from the Swedish Heart Failure Registry.

Beta-blockers reduce mortality and morbidity in heart failure (HF) with reduced ejection fraction (HFrEF). However, patients older than 80 years are poorly represented in randomized controlled trials. We assessed the association between beta-blocker use and outcomes in HFrEF patients aged ≥80 years. We included patients with an ejection fraction <40% and aged ≥80 years from the Swedish HF Registry. The association between beta-blocker use, all-cause mortality and cardiovascular (CV) mortality/HF hospitalization was assessed by Cox proportional hazard models in a 1:1 propensity score-matched cohort. To assess consistency, the same analyses were performed in a positive control cohort with age <80 years. A negative control outcome analysis was run using hospitalization for cancer as endpoint. Of 6562 patients aged ≥80 years, 5640 (86%) received beta-blockers. In the matched cohort including 1732 patients, beta-blocker use was associated with a significant reduction in the risk of all-cause mortality [hazard ratio (HR) 0.89, 95% confidence interval (CI) 0.79-0.99]. Reduction in CV mortality/HF hospitalization was not significant (HR 0.94, 95% CI 0.85-1.05) due to the lack of association with HF hospitalization, whereas CV death was significantly reduced. After adjustment rather than matching for the propensity score in the overall cohort, beta-blocker use was associated with reduced risk of all outcomes. In patients aged <80 years, use of beta-blockers was associated with reduced risk of all-cause death (HR 0.79, 95% CI 0.68-0.92) and of the composite outcome (HR 0.88, 95% CI 0.77-0.99). In HFrEF patients ≥80 years of age, use of beta-blockers was high and was associated with improved all-cause and CV survival.

Chlorhexidine Bathing to Prevent Central Line–Associated Bloodstream Infections in Hematology Units: A Prospective, Controlled Cohort Study

Chlorhexidine (CHG) bathing decreases the incidence of bloodstream infections in intensive care units, but its effect has been understudied in patients with hematological malignancies in noncritical care units. Adults with hematological malignancies hospitalized for cytotoxic chemotherapy in noncritical care units were offered daily 2% CHG bathing. We compared outcomes of patients who chose CHG bathing (CHG group) with outcomes of those who did not choose CHG bathing (usual-care group). The primary outcome was gram-positive cocci-related, skin flora-related, or central line-associated bloodstream infection. The negative control outcome was gut-origin bacteremia. The CHG group (n = 485) had a crude incidence rate of the primary outcome that was 60% lower than the rate for the usual-care group (n = 408; 3.4 vs 8.4 per 1000 patient-days, P = .02) but had a similar crude incidence rate of the negative control outcome (4.5 vs 3.2 per 1000 patient-days; P = .10). In multivariable analyses, CHG bathing was associated with a 60% decrease in the primary outcome (adjusted hazard ratio [HR], 0.4; P < .001). In contrast, CHG bathing had no effect on the negative control outcome (adjusted HR, 1.1; P = .781). CHG bathing was well tolerated by participants in the CHG group. CHG bathing could be a highly effective approach for preventing gram-positive cocci-related, skin flora-related, or central line-associated bacteremia in patients with hematological malignancies who are hospitalized for cytotoxic chemotherapy in noncritical care units.

Early childhood vaccination and subsequent mortality or morbidity: are observational studies hampered by residual confounding? A Danish register-based cohort study

ObjectivesTo estimate the association between childhood vaccination and subsequent morbidity and mortality by adjusting for environmental and host factors. Further, to examine the degree of residual confounding in such observational...

Examining associations between physical activity and cardiovascular mortality using negative control outcomes.

The purpose of a negative control is to reproduce a condition that cannot involve the hypothesized causal mechanism, but does involve the same sources of bias and confounding that may distort the primary association of interest. Observational studies suggest physical inactivity is a major risk factor for cardiovascular disease (CVD), although potential sources of bias, including reverse causation and residual confounding, make it difficult to infer causality. The aim was to employ a negative control outcome to explore the extent to which the association between physical activity and CVD mortality is explained by confounding. The sample comprised 104 851 participants (aged 47 ± 17 years; 45.4% male) followed up over mean (SD) 9.4 ± 4.5 years, recruited from the Health Survey for England and the Scottish Health Survey. There were 10 309 deaths, of which 3109 were attributed to CVD and 157 to accidents (negative control outcome). Accidental death was related to age, male sex, smoking, longstanding illness and psychological distress, with some evidence of social patterning. This confounding structure was similar to that seen with CVD mortality, suggesting that our negative control outcome was appropriate. Physical activity (per SD unit increase in MET-hr-wk) was inversely associated with CVD [hazard ratio (HR) = 0.75; 95% confidence interval (CI), 0.70, 0.80]; the point estimate between physical activity and accidental death was in the same direction but of lesser magnitude (HR = 0.86; 95% CI: 0.69, 1.07). A linear dose-response pattern was observed for physical activity and CVD but not with the negative control. Inverse associations between physical activity and risk of CVD mortality are likely causal but of a smaller magnitude than commonly observed. Negative control studies have the potential to improve causal inference within the physical activity field.

Intraseason Waning of Influenza Vaccine Effectiveness.

In the United States, it is recommended that healthcare providers offer influenza vaccination by October, if possible. However, if the vaccine's effectiveness soon begins to wane, the optimal time for vaccination may be somewhat later. We examined whether the effectiveness of influenza vaccine wanes during the influenza season with increasing time since vaccination. We identified persons who were vaccinated with inactivated influenza vaccine from 1 September 2010 to 31 March 2017 and who were subsequently tested for influenza and respiratory syncytial virus (RSV) by a polymerase chain reaction test. Test-confirmed influenza was the primary outcome and days-since-vaccination was the predictor of interest in conditional logistic regression. Models were adjusted for age and conditioned on calendar day and geographic area. RSV was used as a negative-control outcome. Compared with persons vaccinated 14 to 41 days prior to being tested, persons vaccinated 42 to 69 days prior to being tested had 1.32 (95% confidence interval [CI], 1.11 to 1.55) times the odds of testing positive for any influenza. The odds ratio (OR) increased linearly by approximately 16% for each additional 28 days since vaccination. The OR was 2.06 (95% CI, 1.69 to 2.51) for persons vaccinated 154 or more days prior to being tested. No evidence of waning was found for RSV. Our results suggest that effectiveness of inactivated influenza vaccine wanes during the course of a single season. These results may lead to reconsideration of the optimal timing of seasonal influenza vaccination.

Exposure to ambient air pollution and the incidence of dementia: A population-based cohort study

IntroductionEmerging studies have implicated air pollution in the neurodegenerative processes. Less is known about the influence of air pollution, especially at the relatively low levels, on developing dementia. We conducted a population-based cohort study in Ontario, Canada, where the concentrations of pollutants are among the lowest in the world, to assess whether air pollution exposure is associated with incident dementia. MethodsThe study population comprised all Ontario residents who, on 1 April 2001, were 55–85years old, Canadian-born, and free of physician-diagnosed dementia (~2.1 million individuals). Follow-up extended until 2013. We used population-based health administrative databases with a validated algorithm to ascertain incident diagnosis of dementia as well as prevalent cases. Using satellite observations, land-use regression model, and an optimal interpolation method, we derived long-term average exposure to fine particulate matter (≤2.5μm in diameter) (PM2.5), nitrogen dioxide (NO2), and ozone (O3), respectively at the subjects' historical residences based on a population-based registry. We used multilevel spatial random-effects Cox proportional hazards models, adjusting for individual and contextual factors, such as diabetes, brain injury, and neighborhood income. We conducted various sensitivity analyses, such as lagging exposure up to 10years and considering a negative control outcome for which no (or weaker) association with air pollution is expected. ResultsWe identified 257,816 incident cases of dementia in 2001–2013. We found a positive association between PM2.5 and dementia incidence, with a hazard ratio (HR) of 1.04 (95% confidence interval (CI): 1.03–1.05) for every interquartile-range increase in exposure to PM2.5. Similarly, NO2 was associated with increased incidence of dementia (HR=1.10; 95% CI: 1.08–1.12). No association was found for O3. These associations were robust to all sensitivity analyses examined. These estimates translate to 6.1% of dementia cases (or 15,813 cases) attributable to PM2.5 and NO2, based on the observed distribution of exposure relative to the lowest quartile in concentrations in this cohort. DiscussionIn this large cohort, exposure to air pollution, even at the relative low levels, was associated with higher dementia incidence.

Comparing methods to address bias in observational data: statin use and cardiovascular events in a US cohort.

The theoretical conditions under which causal estimates can be derived from observational data are challenging to achieve in the real world. Applied examples can help elucidate the practical limitations of methods to estimate randomized-controlled trial effects from observational data. We used six methods with varying design and analytic features to compare the 5-year risk of incident myocardial infarction among statin users and non-users, and used non-cardiovascular mortality as a negative control outcome. Design features included restriction to a statin-eligible population and new users only; analytic features included multivariable adjustment and propensity score matching. We used data from 5294 participants in the Cardiovascular Health Study from 1989 to 2004. For non-cardiovascular mortality, most methods produced protective estimates with confidence intervals that crossed the null. The hazard ratio (HR) was 0.92, 95% confidence interval: 0.58, 1.46 using propensity score matching among eligible new users. For myocardial infarction, all estimates were strongly protective; the propensity score-matched analysis among eligible new users resulted in a HR of 0.55 (0.29, 1.05)-a much stronger association than observed in randomized controlled trials. In designs that compare active treatment with non-treated participants to evaluate effectiveness, methods to address bias in observational data may be limited in real-world settings by residual bias.

Non-specific effects of measles, mumps, and rubella (MMR) vaccination in high income setting: population based cohort study in the Netherlands

Objectives To investigate whether measles, mumps, and rubella (MMR) vaccine has positive non-specific effects in a high income setting and to compare rates of hospital admissions for infections between children...

Invited Commentary: Bias Attenuation and Identification of Causal Effects With Multiple Negative Controls.

In this commentary, we describe several extensions to the interesting and important negative control exposure approach for partial confounding adjustment in time-series analysis proposed by Flanders et al. (Am J Epidemiol. 2017;185(10):941-949). Specifically, by leveraging the availability of exposure time series, we show that under certain additional fairly reasonable assumptions, one can incorporate both past and future exposures as multiple negative control exposures to further attenuate confounding bias. We further describe 2 specific settings in which multiple controls can be used to fully account for confounding bias; the first assumes a forward-in-time version of the familiar autoregressive model for the exposure time series, while the second combines a negative control exposure with a negative control outcome for joint indirect adjustment of confounding. We briefly illustrate how one might apply our proposed framework in time-series studies. Both the original method of Flanders et al. and our proposed extensions are particularly well-suited for time-series data such as the air pollution study considered in their paper, and as such should be considered in routine environmental health studies.

A New Method for Partial Correction of Residual Confounding in Time-Series and Other Observational Studies.

Methods exist to detect residual confounding in epidemiologic studies. One requires a negative control exposure with 2 key properties: 1) conditional independence of the negative control and the outcome (given modeled variables) absent confounding and other model misspecification, and 2) associations of the negative control with uncontrolled confounders and the outcome. We present a new method to partially correct for residual confounding: When confounding is present and our assumptions hold, we argue that estimators from models that include a negative control exposure with these 2 properties tend to be less biased than those from models without it. Using regression theory, we provide theoretical arguments that support our claims. In simulations, we empirically evaluated the approach using a time-series study of ozone effects on asthma emergency department visits. In simulations, effect estimators from models that included the negative control exposure (ozone concentrations 1 day after the emergency department visit) had slightly or modestly less residual confounding than those from models without it. Theory and simulations show that including the negative control can reduce residual confounding, if our assumptions hold. Our method differs from available methods because it uses a regression approach involving an exposure-based indicator rather than a negative control outcome to partially correct for confounding.

Habitual coffee consumption and risk of type 2 diabetes, ischemic heart disease, depression and Alzheimer's disease: a Mendelian randomization study.

Observationally, coffee is inversely associated with type 2 diabetes mellitus (T2DM), depression and Alzheimer’s disease, but not ischemic heart disease (IHD). Coffee features as possibly protective in the 2015 Dietary Guidelines for Americans. Short-term trials suggest coffee has neutral effect on most glycemic traits, but raises lipids and adiponectin. To clarify we compared T2DM, depression, Alzheimer’s disease, and IHD and its risk factors by genetically predicted coffee consumption using two-sample Mendelian randomization applied to large extensively genotyped case-control and cross-sectional studies. Childhood cognition was used as a negative control outcome. Genetically predicted coffee consumption was not associated with T2DM (odds ratio (OR) 1.02, 95% confidence interval (CI) 0.76 to 1.36), depression (0.89, 95% CI 0.66 to 1.21), Alzheimer’s disease (1.17, 95% CI 0.96 to 1.43), IHD (0.96, 95% CI 0.80 to 1.14), lipids, glycemic traits, adiposity or adiponectin. Coffee was unrelated to childhood cognition. Consistent with observational studies, coffee was unrelated to IHD, and, as expected, childhood cognition. However, contrary to observational findings, coffee may not have beneficial effects on T2DM, depression or Alzheimer’s disease. These findings clarify the role of coffee with relevance to dietary guidelines and suggest interventions to prevent these complex chronic diseases should be sought elsewhere.

On negative outcome control of unobserved confounding as a generalization of difference-in-differences.

The difference-in-differences (DID) approach is a well known strategy for estimating the effect of an exposure in the presence of unobserved confounding. The approach is most commonly used when pre-and post-exposure outcome measurements are available, and one can assume that the association of the unobserved confounder with the outcome is equal in the two exposure groups, and constant over time. Then, one recovers the treatment effect by regressing the change in outcome over time on the exposure. In this paper, we interpret the difference-in-differences as a negative outcome control (NOC) approach. We show that the pre-exposure outcome is a negative control outcome, as it cannot be influenced by the subsequent exposure, and it is affected by both observed and unobserved confounders of the exposure-outcome association of interest. The relation between DID and NOC provides simple conditions under which negative control outcomes can be used to detect and correct for confounding bias. However, for general negative control outcomes, the DID-like assumption may be overly restrictive and rarely credible, because it requires that both the outcome of interest and the control outcome are measured on the same scale. Thus, we present a scale-invariant generalization of the DID that may be used in broader NOC contexts. The proposed approach is demonstrated in simulations and on a Normative Aging Study data set, in which Body Mass Index is used for NOC of the relationship between air pollution and inflammatory outcomes.

Statin adherence and risk of acute cardiovascular events among women: a cohort study accounting for time-dependent confounding affected by previous adherence

ObjectivesPrevious studies on the effect of statin adherence on cardiovascular events in the primary prevention of cardiovascular disease have adjusted for time-dependent confounding, but potentially introduced bias into their estimates...